On Tue, Aug 28, 2018 at 05:08:25PM -0400, Tom Lane wrote:
> I wrote:
>> Digging, this is because the "make check" processing isn't actually
>> installing oid2name (resp. vacuumlo) into the temp installation tree :-(.
>>
>> I don't think this is the fault of your tests, exactly --- there's
>> something wrong in the existing make rules. I'm not sure what though.
>> EXTRA_INSTALL should be getting set, according to pgxs.mk:
>> temp-install: EXTRA_INSTALL+=$(subdir)
>> but it evidently isn't.
>
> Oh! The reason why not is that that line is inside "ifdef REGRESS",
> and these subdirectories have no REGRESS tests. So we should move
> that out to where it will apply regardless of that. Will fix.
Yes, I was going to send a patch along the same lines as what you have
just committed. Thanks! It looks that the binary was installed in my
own PATH, so I did not notice it.
This should definitely get back-patched in my opinion, I have similar
problems with pg_rewind tests of 9.3 and 9.4 that I still maintain out
of the core tree, where the main binary needs to be installed first. In
this case I just used some Makefile trick.
--
Michael